Effects of part-of-speech, quantity and predictability on acoustic durations in Estonian spontaneous speech

Spontaneous speech is highly variable. When controlling for the number of phones and other predictors, content words are produced with longer durations than function words; more predictable words are produced with shorter durations than less predictable words. These effects further interact with each other such that reduction due to predictability is more extreme for function words than content words. However, these effects have mainly been investigated in English. The current study focuses on Estonian, a Finno-Ugric language with a more complex word class and phonemic quantity system than English. Our analyses indicate that content words (nouns and adjectives) are longer than function words (pronouns and conjunctions) in Estonian, even when controlling for predictors such as the number of phones, speech rate, etc. Whereas words are longer in the second and third quantities than in the first quantity, the effect is not the same for all parts-of-speech. Semantically rich part-of-speech, nouns and adjectives are more affected by the quantity distinction than other parts-of-speech. The duration of both content and function words decreases with increasing predictability. In summary, we show that although the effects of part-of-speech and conditional probability in Estonian are overall similar to English, their exact influence is modulated by characteristic properties of the language in use, such as part-of-speech and quantity distributions.